Testing and Dating Structural Breaks in Generalized Linear Multivariate Models for Stock Market Contagion

The models for testing and dating breaks in stock returns and volatilities often rely on the restrictive assumption of common breaks. This assumption suggests that a shift occurred due to common innovations. Models under this assumption can only be estimated simultaneously. This assumption may be too restrictive because shocks are likely to affect stock returns differently across time. We rely instead on the assumption of distinct breaks and use the sequential testing procedure to test for the distinctiveness of break dates. Using this procedure, we examine if there are breaks in conditional correlations. In addition, we assess the relevance of distinct breaks for estimating models of volatility spillovers. We present evidence of changes in correlations of returns of varying degrees between markets in different regions. We also present evidence of breaks in the conditional correlations and provide evidence of contagion. We detect time-variation in volatility spillovers and find that spillovers are affected by distinct breaks